Study On The Method Of Image Segmentation Based On Markov Random Field

Image segmentation is the key processing technology of image processing and computer vision systems, its accuracy directly affects the accuracy of the process of high-level computer vision system. In recent years, there have been a lot of image segmentation methods. Markov random field model (MRF) has good characteristics of segmentation, based on statistical methods,it is able to make full use of a priori knowledge,few model parameters and easily combining with other methods etc.Therefore,it is widely used in the field of image segmentation.This paper studies image segmentation algorithms based on MRF,and focus on the color image segmentation method based on multiscale MRF in the wavelet domain,and the color texture image segmentation method based on multiscale MRF in the morphological wavelet domain. We do the following works to improve the two methods:Firstly, we discuss the multiscale MRF model in the wavelet domain.In oder to reduce the occurrence of under-segmentation,we introduces a variable weight between multi-spce multi-scale model field and label field to establish their correlation constraints, based on the method that combines multi-space characteristics and makes use of the fuzzy theory to estimate parameter of the model. We demonstrate the validation of our model through the experiments for the color images from the Berkeley database.Secondly, we study the morphological wavelet domain multiscale MRF model.In order to improve the ability of the description of the color texture images, we make full use of the characteristics of RGB space and HSV space to build a multi-space morphological wavelet domain multiscale MRF model. Through the experiments for the color synthesized texture images from the Prague texture database,we demonstrate that our model can gets more consistent regions and improves the accuracy of image segmentation.